SAF apps

Creating Applications with SAF

There are two ways of creating applications using SAF: 1) Deploying a pipeline via the JSON API and 2) Creating custom C++ applications.

1. Running Pipelines with the JSON API

The JSON API allows users to run arbitrary DAGs of cameras and operators using a JSON specification. The pipeline is specified with a pipeline_name and a list of operators. An operator is specified the operator_name, operator_type, a map of the operator's parameters and the name of its inputs. The parameters that a operator takes can be found in the Create(const FactoryParamsType& params) method of operator subclasses. An input name should have the form "<src operator name>:<src sink name>". If <srcSinkName> is output (the default for many operators), you can omit :<srcSinkName> and just list the <src operator Name> under input. Here is an example JSON spec:

{
  "pipeline_name": "FrameWriterExample",
  "operators": [{
      "operator_name": "MyCamera",
      "operator_type": "Camera",
      "parameters": {
        "camera_name": "GST_TEST"
      }
    },
    {
      "operator_name": "MyWriter",
      "operator_type": "FrameWriter",
      "parameters": {
        "format": "binary",
        "output_dir": "/tmp/frame_writer/"
      },
      "inputs": {
        "input": "MyCamera:output"
      }
    }
  ]
}

Deploy the pipeline with pipeline:

/path/to/apps/pipeline/pipeline -p /path/to/pipelines/pipeline.json

This program takes an optional flag -n/--dry-run that will load the pipeline, but not run it. This mode is useful for testing to make sure that the JSON pipeline file is well-formed and that SAF is able to instantiate all of the operators.

If you build SAF with the -DUSE_GRAPHVIZ flag to CMake, then the pipeline app will support the -g/--graph optional flag. This will use GraphViz to visualize the pipeline as a graph (DAG) in an OpenCV window.

2. Custom C++ Applications

Applications can also be made in C++ by adding a directory to the /path/to/saf/apps/. Current existing applications can be used as an example. Building a custom application can be useful when there is processing to be done outside of an operator.

Running Popular SAF Applications

Simple

/path/to/apps/simple
  --camera <camera-id>
  [--display]

Classifier

/path/to/apps/classifier
  --camera <camera-id>
  [--display]
  -m <model-id>

Detector

/path/to/apps/detector
  --camera <camera-id>
  --detector_type [opencv-face|mtcnn-face|yolo|frcnn|ssd|ncs-yolo]
  -m <model-id>
  --detector_confidence_threshold <number>
  --detector_targets <csv-labels>
  [--display]

Tracker

/path/to/apps/tracker
  -c <camera-id>
  --detector_type [opencv-face|mtcnn-face|yolo|frcnn|ssd|ncs-yolo]
  -m <model-id>
  --detector_confidence_threshold <number>
  --detector_targets <csv-lables>
  [--display]
  --tracker_type [dlib]
  --detector_idle_duration <secs>
  --tracker_calibration_duration <secs>